Accurately constraining velocity information from spectral imaging observations using machine learning techniques: Fitting velocities with machine learning

Determining accurate plasma Doppler (line-of-sight) velocities from spectroscopic measurements is a challenging endeavour, especially when weak chromospheric absorption lines are often rapidly evolving and, hence, contain multiple spectral components in their constituent line profiles. Here, we present a novel method that employs machine learning techniques to identify the underlying components present within observed spectral lines, before subsequently constraining the constituent profiles through single or multiple Voigt fits. Our method allows active and quiescent components present in spectra to be identified and isolated for subsequent study. Lastly, we employ a Ca II 8542 {\AA} spectral imaging dataset as a proof-of-concept study to benchmark the suitability of our code for extracting two-component atmospheric profiles that are commonly present in sunspot chromospheres. Minimisation tests are employed to validate the reliability of the results, achieving median reduced $\chi^2$ values equal to 1.03 between the observed and synthesised umbral line profiles.